This article is auto-synced from its in-app version in Tai.

Ten complete first messages for the Connector Builder, from keyless public APIs to authenticated SaaS tools, each with a note on what the resulting connector does.

When you'd use this

You want to see the whole build flow end to end before describing your own source, or you need a starting prompt for a source similar to one below. Each prompt is a complete first message. Paste it into a fresh Builder chat, approve the plan, and enter credentials where noted. The keyless sources are the fastest way through it.

Weather forecast (Open-Meteo)

Connect https://open-meteo.com/en/docs. Agents should answer "what's the
weather in <city> on <date>?", so they need to turn a city name into
coordinates first, then fetch a daily forecast (min/max temperature,
precipitation, weather code) for up to 14 days. No key needed.

What you get: two operations, geocode a place name and fetch the daily forecast for coordinates. Two hosts, the geocoding host and the forecast host, both declared in the plan. No credentials.

Exchange rates (Frankfurter)

Connect https://frankfurter.dev. Agents should answer "what is 1 EUR in
CHF today?", "what was the USD/GBP rate on <date>?", and "which currencies
are supported?". Keyless.

What you get: latest rates, rates on a given date, and the currency list. Rates come from the European Central Bank once per working day, so weekends and holidays return the last working day. Leave the conversion arithmetic to the agent.

Country facts (REST Countries)

Connect https://restcountries.com (v3.1). Agents should answer basic facts
about one country: capital, region, population, currencies, languages,
calling code, driving side, timezones. Look up by name or ISO code. Only
return those fields; the full record is large.

What you get: a single lookup operation with a trimmed field list. A good example of shaping: the full country record is huge, and agents only need a dozen fields.

Public holidays (Nager.Date)

Connect https://date.nager.at/Api. Agents should answer "is <date> a
public holiday in <country>?", "list this year's holidays in <country>",
and "what are the next long weekends in <country>?". Keyless, ISO country
codes.

What you get: holidays for a country and year, plus the long-weekend operation. Around a hundred countries are covered.

Encyclopedia summaries (Wikipedia)

Connect the Wikipedia REST API (https://en.wikipedia.org/api/rest_v1/).
Agents should answer "give me a short summary of <topic>" with the extract,
a thumbnail URL and the page URL. English Wikipedia only.

What you get: one operation, page summary by title. Another language edition or a sister project such as Wikivoyage is a separate host, so a separate connector.

National parks (NPS API)

Connect the US National Park Service API (https://www.nps.gov/subjects/developer/api-documentation.htm).
Agents should answer "which national parks are in <state>?" and "are there
current alerts or closures for <park>?". It needs an API key sent as the
X-Api-Key header. I'll enter it in the Credentials panel.

What you get: parks by state code and alerts by park code. The first authenticated example: the assistant declares the api_key field, you save the value, and self-tests run against the real API.

Helpdesk conversations (Intercom)

Connect the Intercom REST API (https://developers.intercom.com/docs/references/rest-api/api.intercom.io/).
Our workspace is hosted in the EU, so use api.eu.intercom.io. Agents should
answer "which open conversations mention <keyword>?" and "what happened in
conversation <id>?". Return the contact, state, subject and the message
bodies. Auth is a Bearer access token in the Authorization header.

What you get: a conversation search plus one-conversation detail. Intercom's search is an HTTP POST that only reads, so the assistant treats it as a read operation. Save the token as Bearer followed by a space and the token.

Support tickets (HubSpot)

Connect HubSpot's CRM API (https://developers.hubspot.com/docs/api/crm/tickets).
Scope: ONLY Service Hub tickets, no contacts, companies or deals. Agents
should answer "which tickets are open for <company or email>?", "what is
the status and priority of ticket <id>?", and "what changed on it recently?".
Auth is a private-app token as Bearer in the Authorization header.

What you get: a focused ticket connector from a very large API. The scope line is what keeps the plan to a few operations. Save the token as Bearer followed by a space and the token.

Project board (Monday.com)

Connect Monday.com's GraphQL API (https://developer.monday.com/api-reference).
Only board <board id>. Agents should answer "which items are assigned to
<person>?", "what is the status of <item>?", and "which items are overdue?".
Auth is the API token in the Authorization header.

What you get: read-only GraphQL queries sent as POST requests. Monday returns errors inside the body with an HTTP 200, so the assistant checks the response body rather than the status code. One board per connector keeps the queries simple.

Spreadsheet-style records (Airtable)

Connect the Airtable API (https://airtable.com/developers/web/api/introduction)
for base <base id>, table "<table name>". Agents should answer "which
records match <condition>?" and "show record <id>" with these fields:
<list the fields>. Use a read-only personal access token as Bearer in the
Authorization header.

What you get: filtered listing with pagination, plus a single-record lookup. Create a token with read-only scope for exactly this base.

Patterns worth reusing

Across the examples, the prompts that worked share the same shape:

Name the source and link its documentation. The assistant reads the docs to design the plan; a link saves a round of questions.
State the questions agents should answer. Operations fall out of the questions.
For a big API, decide the scope. Say what is in and, just as important, what is out.
Name the auth mechanism and where the value goes. Header, query parameter or Basic auth. The value itself goes in the Credentials panel, never in the chat.
Name the fields that matter. That is what the assistant uses to shape results.

Tips and limits

Running any of these prompts in the Builder requires the Company Admin role; browsing the resulting connectors is open to all users.
GraphQL sources are read-only, like every other connector. The platform reads each query before it is sent and refuses anything that would write or subscribe, so ask for queries and leave changes to the tool itself.
The prompts reflect each source's public documentation at the time of writing. If a source has changed its API, the assistant notices during the self-test and adjusts the plan, or asks you.
Authenticated examples need a key or token from your own account with that tool. Enter it in the Credentials panel, never in the chat.
Every example builds a read-only connector. Anything that creates, updates or deletes records in the source is out of scope.

Connector Builder examples